My research asks how humans (and in particular babies) build internal models of structured environments: from infants discovering the regularities of their mother tongue, to adults accumulating evidence towards a decision, to the temporal structure that shapes what we perceive. I approach these questions along three complementary strands.

Strand 01

Statistical and network learning in sequences

How listeners, even sleeping newborns, extract transition probabilities and the graph structure hidden in streams of sound.

EEG cap on a sleeping neonate beside a diagram of long-horizon associative strength between sequence elements
Long-horizon associative learning: every element is bound to its whole recent history with an exponentially decaying weight, here measured with high-density EEG in sleeping neonates.

A central question of my PhD: how do humans extract the latent structure of sound sequences? We have shown that sleeping neonates (two or three days old) can learn transition probabilities between syllables and use that information to segment a continuous speech stream (Benjamin et al., Dev. Sci., 2023; Fló et al., Sci. Rep., 2022).

Beyond local transition probabilities, we found that people prune and complete the underlying network structure of the input (Benjamin et al., eLife, 2023), and that long-horizon associative learning unifies local transition-probability learning with high-order graph learning in a single mechanism (Benjamin et al., J. Neurosci., 2024; Benjamin et al., PNAS, 2026).

With colleagues at Aix-Marseille University (J. Pesnot-Lerousseau and B. Morillon) and in Quebec (P. Albouy), I am now using intracranial recordings to resolve the fine-grained neural basis of this learning in humans.

Strand 02

Decision making and sensory integration

How temporal prediction, evidence accumulation and rule discovery interact while the brain commits to a choice.

MEG decoding time courses for stimulus aspect, evidence and belief update, with their cortical source maps
MEG decoding of three latent variables during perceptual decisions: stimulus aspect, accumulated evidence and belief update, together with their cortical sources.

In my postdoc with B. Morillon and V. Wyart, we asked how temporal predictions, sensory integration and rule discovery interact during perceptual decisions, and found them to be interdependent rather than separable inference processes (Benjamin et al., PNAS, 2026).

My current project (work in progress) combines MEG with computational models that bridge Bayesian inference and recurrent neural networks, to investigate the neural basis of flexible evidence accumulation and the role abstraction plays in it.

Strand 03

Auditory cortex development and prematurity

How early auditory experience sculpts the superior temporal sulcus in the newborn brain.

MRI scanner icon beside curves of right superior temporal sulcus depth as a function of birth term
Depth of the right superior temporal sulcus at term-equivalent age: the earlier the birth, the shallower the sulcus: an anatomical trace of early auditory experience.

With G. Dehaene-Lambertz and the Geneva neonatal team, we used MRI to study how early auditory experience shapes the structure of the superior temporal sulcus in newborns (Benjamin et al., Brain Struct. Funct., 2025), and how gestational age and sex affect its functional connectivity at term-equivalent age (Mancuso et al., Brain Struct. Funct., 2025).

Toolkit

Methods

Measuring the signal, then explaining it with a model that could have produced it.

  • BehaviourPsychophysics and sequence-learning tasks in adults, infants and patients.
  • EEG · MEG · iEEGHigh-density recordings, including neonatal EEG and intracranial recordings in epileptic patients.
  • MRI · fMRIAnatomical morphometry of the neonatal brain and task-based functional imaging.
  • Computational modelsMathematical accounts of behaviour, Bayesian observers, recurrent neural networks.

Path

Background

From signal processing and bio-engineering to the developing brain.

  • 2024 → Postdoctoral researcher INS, Aix-Marseille Université (B. Morillon) and LNC², ENS-PSL (V. Wyart). Funded by the Fondation pour la Recherche Médicale.
  • 2023 PhD in cognitive neuroscience NeuroSpin (CEA & Sorbonne Université), supervised by Ghislaine Dehaene-Lambertz, on learning temporal dependencies in auditory sequences, in adults and neonates.
  • 2018–2019 Research visits University College London (M. Chait, auditory salience and pupillometry) and the Montreal Neurological Institute, McGill (R. Zatorre, P. Albouy, B. Morillon, speech and music processing).
  • 2015–2019 Engineering MSc & MSc in computational biology CentraleSupélec, major in bio-engineering and signal processing, with a parallel MSc at Université Paris-Saclay.